Research map / Video / Action recognition
HAA500: Human-Centric Atomic Action Dataset With Curated Videos
Jihoon Chung, Cheng-hsin Wuu, Hsuan-ru Yang, Yu-Wing Tai, Chi-Keung Tang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add atomic action labels to composite labels in training (ablation) | I3D RGB · HAA500 Instrument composite classification | top-1 accuracy: 70.59% -> 71.90% | Improved |
| add atomic action labels to composite labels in training (ablation) | I3D RGB · HAA500 Sport composite classification | top-1 accuracy: 47.48% -> 53.93% | Improved |
| add atomic action labels to composite labels in training (ablation) | TSN RGB · HAA500 Sport composite classification | top-1 accuracy: 72.64% -> 72.48% | Hurt |
| add pose input to RGB (ablation) | I3D · HAA500 | top-1 accuracy: 33.53% -> 42.80% | Improved |
| add pose input to RGB (ablation) | I3D · Gym288 | top-1 accuracy: 76.11% -> 77.31% | Improved |
| pre-train on HAA500 instead of no pre-training | I3D-RGB · UCF101 fine-tuning | top-1 accuracy: 58.87% -> 68.70% | Improved |
| pre-train on HAA500 instead of no pre-training | I3D-RGB · HMDB51 fine-tuning | top-1 accuracy: 28.56% -> 40.45% | Improved |
| pre-train on AVA instead of no pre-training | I3D-RGB · UCF101 fine-tuning | top-1 accuracy: 58.87% -> 48.54% | Hurt |
| use relaxed (loosely trimmed) clips instead of precise temporal annotation (ablation) | HAA500 · HAA500 test | top-1 accuracy: 33.53% -> 22.80% | Hurt |
| use relaxed HAA500 for pre-training (ablation) | I3D-RGB · UCF101 fine-tuning | top-1 accuracy: 68.70% -> 62.24% | Hurt |
| normalize frames by cropping person-of-interest (ablation) | I3D-RGB · HAA500 Instrument composite classification | top-1 accuracy: 66.01% -> 75.82% | Improved |
| add object detection heatmap channels to RGB (ablation) | I3D-RGB · HAA500 | top-1 accuracy: 33.53% -> 33.73% | No clear effect |
Closest papers by meaning
- AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions (CVPR 2018 · similarity 0.89)
- A Hierarchical Pose-Based Approach to Complex Action Understanding Using Dictionaries of Actionlets and Motion Poselets (CVPR 2016 · similarity 0.89)
- Rethinking Zero-Shot Action Recognition: Learning from Latent Atomic Actions (ECCV 2022 · similarity 0.88)
- UCF-STAR: A Large Scale Still Image Dataset for Understanding Human Actions (AAAI 2020 · similarity 0.88)
- H-MoRe: Learning Human-centric Motion Representation for Action Analysis (CVPR 2025 · similarity 0.86)
- Action-slot: Visual Action-centric Representations for Multi-label Atomic Activity Recognition in Traffic Scenes (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
